Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Task Learning with Feature-Similarity Laplacian Graphs for Predicting Alzheimer's Disease Progression

Zixiang Xu, Menghui Zhou, Jun Qi, Xuanhan Fan, Yun Yang, Po Yang

TL;DR

Alzheimer's disease progression prediction from longitudinal MRI data is challenged by static feature-sharing assumptions. The authors introduce the Feature Similarity Laplacian (FSL) penalty within a Multi-Task Learning (MTL) framework, solved via ADMM, to model evolving feature relationships over time. On the ADNI dataset, MTL-FSL achieves state-of-the-art predictions for ADAS and MMSE and identifies temporally stable MRI biomarkers, enhancing both accuracy and interpretability. This approach advances longitudinal neurodegeneration analysis and can be extended to other diseases and multi-modal data sources.

Abstract

Alzheimer's Disease (AD) is the most prevalent neurodegenerative disorder in aging populations, posing a significant and escalating burden on global healthcare systems. While Multi-Tusk Learning (MTL) has emerged as a powerful computational paradigm for modeling longitudinal AD data, existing frameworks do not account for the time-varying nature of feature correlations. To address this limitation, we propose a novel MTL framework, named Feature Similarity Laplacian graph Multi-Task Learning (MTL-FSL). Our framework introduces a novel Feature Similarity Laplacian (FSL) penalty that explicitly models the time-varying relationships between features. By simultaneously considering temporal smoothness among tasks and the dynamic correlations among features, our model enhances both predictive accuracy and biological interpretability. To solve the non-smooth optimization problem arising from our proposed penalty terms, we adopt the Alternating Direction Method of Multipliers (ADMM) algorithm. Experiments conducted on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that our proposed MTL-FSL framework achieves state-of-the-art performance, outperforming various baseline methods. The implementation source can be found at https://github.com/huatxxx/MTL-FSL.

Multi-Task Learning with Feature-Similarity Laplacian Graphs for Predicting Alzheimer's Disease Progression

TL;DR

Alzheimer's disease progression prediction from longitudinal MRI data is challenged by static feature-sharing assumptions. The authors introduce the Feature Similarity Laplacian (FSL) penalty within a Multi-Task Learning (MTL) framework, solved via ADMM, to model evolving feature relationships over time. On the ADNI dataset, MTL-FSL achieves state-of-the-art predictions for ADAS and MMSE and identifies temporally stable MRI biomarkers, enhancing both accuracy and interpretability. This approach advances longitudinal neurodegeneration analysis and can be extended to other diseases and multi-modal data sources.

Abstract

Alzheimer's Disease (AD) is the most prevalent neurodegenerative disorder in aging populations, posing a significant and escalating burden on global healthcare systems. While Multi-Tusk Learning (MTL) has emerged as a powerful computational paradigm for modeling longitudinal AD data, existing frameworks do not account for the time-varying nature of feature correlations. To address this limitation, we propose a novel MTL framework, named Feature Similarity Laplacian graph Multi-Task Learning (MTL-FSL). Our framework introduces a novel Feature Similarity Laplacian (FSL) penalty that explicitly models the time-varying relationships between features. By simultaneously considering temporal smoothness among tasks and the dynamic correlations among features, our model enhances both predictive accuracy and biological interpretability. To solve the non-smooth optimization problem arising from our proposed penalty terms, we adopt the Alternating Direction Method of Multipliers (ADMM) algorithm. Experiments conducted on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that our proposed MTL-FSL framework achieves state-of-the-art performance, outperforming various baseline methods. The implementation source can be found at https://github.com/huatxxx/MTL-FSL.

Paper Structure

This paper contains 16 sections, 1 theorem, 17 equations, 4 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Multi-task Learning
  5. Feature correlation
  6. Feature correlation
  7. Feature Correlation Fusion Matrix
  8. Feature Similarity Laplacian graph Penalty
  9. One Novel MTL Framework
  10. Optimization Algorithms
  11. Experimental Result
  12. Model Training and Evaluation
  13. Comparison of Models
  14. Longitudinal Prediction Performance
  15. The Identification of Longitudinal MRI Biomarkers
  16. ...and 1 more sections

Key Result

Lemma 1

For any $\lambda_1 \geq 0$, we can calculate Eq.eq:update_Q by the following: where $q_{ij}^{(t+1)}$ and $\theta_{ij}^{(t+1)}$ denote the element of matrix $Q^{(t+1)}$ and $\Theta^{(t+1)}$ respectively.

Figures (4)

  • Figure 1: The overall architecture of the Feature Similarity Laplacian graph Multi-Task Learning framework. The framework consists of two main components: a Feature Similarity Learning module that constructs a feature similarity graph, and a Multi-Task Learning module that leverages this graph to improve prediction performance.
  • Figure 2: The convergence situation of using ADMM algorithm in the ADAS and MMSE datasets.
  • Figure 3: The average root Mean Squared Error (rMSE) compared to the baseline model is presented.
  • Figure 4: The stability selection results of the ADAS and MMSE prediction tasks.

Theorems & Definitions (1)

  • Lemma 1